What the EU AI Act requires (Article 50)
Effective August 2, 2026, Article 50 of the EU AI Act mandates clear labeling of AI-generated or AI-manipulated content. The rule applies to any provider or deployer whose content reaches EU users — including non-EU companies, individual creators, bloggers, and businesses.
What must be labeled
- Text generated by AI if it could be mistaken for human-written (news articles, op-eds, descriptions)
- Photorealistic AI images that could be mistaken for real photos
- AI-edited images where the edit changes meaning (object removal, background replacement, person addition)
- Synthetic voices generated by AI (TTS, voice cloning)
- AI-generated video including talking-head, animation, or photorealistic footage
- Deepfakes of real people — strictest labeling requirements
What's exempt
- Obviously stylized art (cartoons, abstract images) — no realism risk
- AI-edited photos where the edit is minimal (color correction, sharpening)
- Spam filters, classifiers, recommendation engines (not "content")
- Content used in confidential professional settings (medicine, security)
What a compliant label needs
- Human-readable text — "Generated with AI", "AI-generated", "Created by AI", or equivalent
- Visual icon or marker — a clear AI/robot/sparkle symbol or the word "AI"
- Machine-readable metadata — schema.org, C2PA, or XMP metadata for crawlers
- Visibility — must be visible without action by the user (no hover-only tooltips)
- Accessibility — readable contrast (4.5:1 minimum), screen-reader compatible
This generator produces all four: an HTML inline badge, an image overlay snippet, plain-text wording, and a JSON-LD schema block. Drop them into your site and you're compliant.
Where to put the label
- For text articles: at the top, before the headline, OR right under the byline
- For images: bottom-left corner overlay (use the image overlay snippet) AND alt text mention
- For video: first 3 seconds + lower-third throughout + description
- For audio: first 3 seconds verbal disclosure + description text
- For social posts: first line of caption, before any other text
Penalties for non-compliance
Up to €15,000,000 or 3% of total worldwide annual turnover (whichever is higher) for the most serious violations. Smaller violations: €7.5M or 1% of turnover. The fines apply to non-EU companies with any EU customer reach.
FAQ
Do I need this for AI tools that just edit metadata or compress images? No. The rule applies to generated or materially manipulated content that could mislead viewers about authenticity.
I'm in the US — does this apply to me? Yes if you have any EU customer or visitor. The EU AI Act has extraterritorial reach, similar to GDPR. Many US states (NY, California, Texas) have similar laws coming online in 2026-2027.
Are there equivalent US laws yet? California's AB-1836 (effective 2025) and AB-2655 require labeling for political deepfakes. New York's AI disclosure law for media advertising effective June 2026. Federal law (No FAKES Act) pending.
What about AI-assisted writing where I edited heavily? The rule kicks in when AI did the substantive generation. Light editing of AI output still requires disclosure. Heavy human rewriting where the AI was a brainstorm/draft tool is generally exempt.